- Thành phố Hồ Chí Minh, Hồ Chí Minh, Vietnam
- On-site
- Full-time · Mid
- 3+ years
Build and operate the real-time interaction layer that makes our agents sound and behave like great human callers: fast, natural, and reliable under production load.
Responsibilities
- Own the real-time orchestration of the cascade pipeline (ASR → LLM → TTS)
- Build and maintain the real-time voice components: real-time speech understanding, real-time speech generation, voice activity and end-of-turn detection, barge-in and interruption handling
- Drive end-to-end latency down across the interaction layer, from audio in to first audio out
- Integrate and switch between ASR/TTS vendors and self-hosted speech models behind clean, well-tested interfaces
- Deploy and scale self-hosted speech models on GPU (batching, streaming inference, cost per call)
Requirements
- Nice to have: vLLM
- Nice to have: Triton
- Nice to have: TensorRT
- Nice to have: SIP/telephony concepts
- Nice to have: speech-to-speech models
